Skip to content

Use cases

When you run more agents than you can watch

Parallelism breaks down at the interface, not the agent. Four threads means four places something can be waiting. Helicon puts every thread in one sidebar grouped by working directory, marks which are running and which are blocked on an approval, and shows what the fan out cost per thread.

Last updated · Markdown version

Isolation first

Two agents on one checkout will fight. Give each task an isolated git worktree, add each as a project, and the collisions stop being possible rather than being unlikely.

git worktree add ../proj-auth -b feat/auth
git worktree add ../proj-search -b feat/search

Then triage

SignalWhat you do
Thread marked runningLeave it
Thread marked blockedOpen it, read the request, decide
Thread finishedReview the diffs in the thread
Tool call stuckBackground it, stop it, or stop everything

Then count the cost

Fan out spends tokens faster than sequential work, which is the whole point and also the risk. The usage page breaks cost down by thread, by day and by model, so the tradeoff between wall clock time and spend is visible instead of assumed.

Frequently asked questions

How many threads is too many?

The limit is usually your attention rather than the machine. The sidebar exists so the number you can supervise is higher than the number you can remember.

Do worktrees appear separately?

Yes, each as its own project with a badge, because each is its own working directory.

Can I stop everything at once?

Yes, and it ends running calls without ending the sessions.

Same Muse Code. Same subscription. Better interface. 

Free and MIT licensed. Signed Windows installer, universal macOS DMG, and a web build against a daemon you run.